Research On Attribute Reduction Based On Granularity Of Knowledge In Incomplete Knowledge Systems And Its Application

Rough Set Theory, introduced by Professor Pawlak in 1982, is a theory to analyze data. Its basic idea goes as follows: Based on the viewpoint that classification capability is maintained invariably, it proposes the model of kownledge presentation in the basis of Indiscernibility Relation , and uses Upper Approximation and Lower Approximation to approach unprecise objects. Also this theory endues the kownledge with clear mathematics significant, presents kownledge reduction, and derives the decision and classification rule.People use language to ponder, judge, and make inference, but the language is a very thick granularity. How to carry on inferences and judgment using the language? The answer is "To use word (granularity) to calculate". The Granularity-Calculate-based methods of decision-making, reasoning and the recognition are the methods mostly drawing close to humanity's thought to solve the question, and it has the broad application prospect to the complicated proposal of information system. The classical rough set theory proposed that knowledge is granular, but it does not quantify this granular knowledge. This paper proposes definitions of granular amount of knowledge, granular particle amount of knowledge, average granular particle amount of kowledge , approach 1 coefficient and granularity of knowledge; It introduces a method of reduction of knowledge to avoid blindness when choosing the reduction of attributes; Also,the time complexity is analyzed; Finally,the validity is shown through an example of maritime accidents.Granularity Calculation is a hot field in information science. And data with uncertainty and fuzziness in the real life is common. Towards incomplete knowledge systems, this paper proposes definitions of Granularity Interactive Information Degree, which can well describe the infection of decision attribute on condition attributes when attribute-reduction is carried on. This paper takes Granularity Interactive Information Degree as the choosing standard to definite a new Attribute-Significance-Degree formula, and finally proposes a kind of new attribute-reduction algorithm. This algorithm has the strong anti-noise ability.